AS a UMTS enhancement function, High Speed Downlink

Transcription

1 Energy-Efficient Channel Quality ndication (CQ) Feedback Scheme for UMTS High-Speed Downlink Packet Access Soo-Yong Jeon and Dong-Ho Cho Dept. of Electrical Engineering and Computer Science Korea Advanced nstitute of Science and Technology (KAST) Guseong-dong Yuseong-gu Daejeon, Republic of Korea TEL: , FAX: and Abstract A major evolution of UMTS standard is the High Speed Downlink Packet Access (HSDPA). Several new technologies are applied to HSDPA system. One of key techniques supporting HSDPA is the adaptive modulation and coding (AMC) in which the modulation scheme and the coding rate are adaptively changed according to the downlink channel quality reported by the user equipment (UE). Therefore, the channel quality indication (CQ) feedback scheme is directly related to the accuracy of AMC and the performance of HSDPA. This paper proposes an enhanced CQ feedback scheme that can be used when a proportional fair scheduling algorithm (PFA) is used as a packet scheduling algorithm. The enhanced CQ feedback scheme uses a dynamic threshold to filter off redundant CQ feedbacks. With the proposed scheme, the battery capacity of UE can be conserved, maintaining the performance of traditional CQ feedback scheme.. NTRODUCTON AS a UMTS enhancement function, High Speed Downlink Packet Access (HSDPA) was specified in 3rd generation partnership project (3GPP) to support high data rate. HSDPA is designed to reuse existing UMTS structure and functionality as much as possible and intended for low to medium user speed and urban/indoor scenarios. Several new technologies such as adaptive modulation and coding (AMC), hybrid automatic repeat request (HARQ), etc. are applied to HSDPA system. With the AMC technique, the modulation scheme and the coding rate are adaptively changed according to the CQ reported by the UE. So, exact channel quality estimation through a proper CQ feedback scheme is essential in AMC. n current specification of HSDPA, each UE must periodically report the CQ indicating the downlink channel condition to Node B. The CQ value is transmitted by using specific physical uplink channel HS-DPCCH. Clearly, the short feedback cycle of CQ gives better throughput performance than the long feedback cycle of CQ, since the short feedback cycle of CQ enables Node B to estimate downlink channel condition exactly. However, the short feedback cycle of CQ gives larger CQ feedback overhead and uplink interference than the long feedback cycle of CQ. This research was supported in part by University T Research Center Project. n this paper, we propose an enhanced CQ feedback scheme that can be used when a proportional fair scheduling algorithm (PFA) is used as a packet scheduling algorithm. With the proposed scheme, the number of redundant CQ feedbacks is decreased, maintaining the performance of traditional CQ feedback scheme. As a result, the overhead of CQ feedback is reduced, and hence the battery capacity of UE is conserved and the uplink interference is lowered. The rest of this paper is organized as follows: Section introduces the PFA, and section introduces the CQ feedback scheme in 3GPP specification and proposes the enhanced CQ feedback scheme. n section V, we describe the simulation environments and evaluate the performance of the proposed scheme and section V analyzes the uplink interference. Finally, the last section concludes this study.. PROPORTONAL FAR SCHEDULNG ALGORTHM To share channel resource evenly and maximize throughput, an proportional fair scheduling algorithm (PFA) was proposed. With the PFA, the scheduler transmits data to the UE that has the highest proportional fairness priority. The proportional fairness priority of UE i at time t, P UEi (t), is calculated as follows P UEi (t) =DRC i (t)/r i (t) (1) where DRC i (t) is data rate of UE i at time t, determined by Node B according to its downlink channel condition in a given slot, and R i (t) is average data rate received by the UE i over afixedwindowsize(t c ). Average data rate of UE i, R i (t), is updated as follows R i (t +1)=(1 1/t c )R i (t)+1/t c Current T ransmission Rate of UE i (2) where t c is 1000 [1]. Current T ransmission Rate of UE i is equal to DRC i (t), ifue i is served at t. Otherwise, it is /05/$ EEE 245

2 . 3GPP SPECFED AND ENHANCED CQ FEEDBACK SCHEME A. Periodic CQ Feedback Scheme A periodic CQ feedback scheme is on the specification Release 5 of HSDPA. The definition of the CQ and UE procedure for CQ feedback are described in [2]. According to the specification, each UE measures downlink channel quality and selects the suitable CQ values which indicate transport block size, number of HS-PDSCH codes and modulation schemes. The selected CQ values must be the one that the transport block error probability would not exceed 0.1 under the measured downlink channel condition. After selection of the suitable CQ value, UE reports the selected CQ value to Node B at its own CQ feedback timing. The CQ feedback timing is determined by each UE s connection frame number (CFN) given at the radio link establishment procedure. The CQ value is transmitted by using specific physical channel HS-DPCCH. The frame structure of HS-DPCCH is described in Fig. 1 [3]. The CQ value consists of 5 bits and the field size of CQ value on HS-DPCCH is 20 bits. Each UE periodically reports the CQ value to Node B. The feedback cycle of CQ is defined in [4]. According to the specification, the feedback cycle of CQ is defined as 0, 1, 2, 4, 5, 10, 20, 40, 80 subframes, where the subframe length is 2ms. The feedback cycle of CQ is informed to each UE by higher layer signalling. Clearly, as the feedback cycle of CQ is shorter, the throughput performance is better, since the short feedback cycle of CQ enables exact downlink channel estimation of Node B, but the overhead of CQ and uplink interference are larger. Conversely, as the feedback cycle of CQ is longer, the overhead of CQ and the uplink inference are smaller, but the throughput performance is poorer. n addition to the periodic CQ feedback scheme, activitybased CQ feedback scheme and NACK-based CQ feedback scheme are proposed in [5][6]. n the activity-based CQ feedback scheme, an additional CQ feedback is sent with every ACK or NACK. n the NACK-based CQ feedback scheme, an additional CQ feedback is sent with every NACK. Fig. 2 shows the diagram illustrating activity-based CQ feedback and NACK-based CQ feedback. Both schemes enable more exact downlink channel estimation of Node B using extra CQ feedbacks. However, both schemes use more power to send additional CQ feedbacks. After receiving the CQ value from each UE, Node B selects the suitable modulation scheme, coding rate and number of codes for HS-PDSCH, where HS-PDSCH is downlink physical channel used to transmit the data. n order to provide control information to each UE in time, Node B uses HS-SCCH. Each UE must monitor maximum 4 HS-SCCH simultaneously. By monitoring HS-SCCH, each UE is informed of the upcoming data transmission, and could prepare to receive data. B. Enhanced CQ Feedback Scheme An enhanced CQ feedback scheme is proposed to allow effective scheduling and MCS selection for transmissions on the HS-PDSCH, while at the same time minimizing the uplink 246 Fig. 1. Frame structure for uplink HS-DPCCH Fig. 2. Diagram illustrating (a) Activity-based CQ feedback and (b) NACKbased CQ feedback interference and battery power consumption due to CQ transmissions on HS-DPCCH. The enhanced CQ feedback scheme is based on the Release 5 specifications. The specification related to the method of deriving an individual CQ value by the UE is not changed in the enhanced CQ feedback scheme. The enhanced CQ feedback scheme can be used when PFA is used as a packet scheduling algorithm. Since PFA schedules the UEs according to the proportional fairness priority of UEs, the UEs which have low proportional fairness priorities would be less likely scheduled for transmission. Since the CQ value is the decision parameter related to transmission on HS-PDSCH, the CQ feedbacks of UEs which have low proportional fairness priorities can be considered somewhat redundant. Motivated by this idea, a dynamic threshold is introduced to filter off the CQ feedbacks of UEs which have low proportional fairness priorities. n the enhanced CQ feedback scheme, Node B informs threshold value, Θ, to each UE through HS-SCCH. Since each UE must be able to monitor maximum 4 HS-SCCH simultaneously by the specification, one HS-SCCH can be used to inform Θ to each UE. Each UE calculates its own proportional fairness priority, and reports the CQ value to Node B at its own CQ feedback timing, only when its proportional fairness priority is higher than Θ. The threshold

3 value, Θ, is calculated by Node B as follows. Θ=S(,T load ) P last (3) TABLE SMULATON PARAMETERS where S(,T load ) is scaling function according to the feedback cycle of CQ,, signalled by higher layer and current traffic load, T load. P last is the proportional fairness priority of UE served by Node B in last subframe. When the number of UE served in last subframe is more than one, P last is set to the lowest one among the proportional fairness priorities of UEs served, and when no one is served in last subframe, P last is set to zero. f Θ is zero, the enhanced CQ feedback scheme is the same as the periodic CQ feedback scheme, since no CQ feedbacks are filtered off. With the enhanced CQ feedback scheme, the number of redundant CQ feedback can be decreased, and hence the battery capacity of UE is saved and uplink interference is reduced, since the UE that is less likely scheduled for transmission due to the low proportional fairness priority would not transmit the CQ to Node B. However, if the threshold value, Θ, is not set properly in the enhanced CQ feedback, too many UEs that have lower proportional fairness priority than Θ may not transmit the CQ to Node B, so the performance such as throughput, delay and fairness may be poorer than that with the periodic CQ feedback scheme. Therefore, to adjust Θ properly, S(,T load ) is introduced, which is the value between 0 and 1. S(,T load ) adjusts Θ to achieve following performance criterions. T hroughput criterion : 98% throughput of the periodic CQ feedback scheme Delay criterion : no more than 10% increase in view of delay of the periodic CQ feedback scheme Fairness criterion : no more than 5% decrease in view of fairness of the periodic CQ feedback scheme The throughput and delay are clear measures to compare. To provide a numerical measure reflecting the fairness, we use the jain fairness index (F jain ) defined as follows [7]. F jain =( b i ) 2 /(N b 2 i ) (4) where N is the number of UEs, and b i is the throughput of UE i. The jain fairness index is a value between 0 and 1, and as the system provides more fair service, the jain fairness index is closer to 1. The value of S(,T load ) satisfying these criterions can be found through the repeated simulations. V. SMULATON CONFGURATON AND RESULTS The default simulation parameters are as summarized in Table. We consider 5 MCS levels shown in Table and the link layer simulation is executed using the FER data in [9]. Fig. 3 shows the scaling factor, S(,T load ), according to the feedback cycle of CQ,, and traffic load, T load. With S(,T load ) in Fig. 3, the enhanced CQ feedback scheme gives maximum battery saving of UEs, satisfying the throughput, delay and fairness criterions. S(,T load ) in Fig. 3 is founded 247 Parameter Explanation / Assumption Cellular layout Hexagonal 19 cells, no sector Site to site distance 2800 m Antenna pattern Omni Propagation model Proportional to 1/d 4, d in meters Std. deviation of slow fading 8dB Fast fading 1 path rayleigh Carrier frequency 2.19 GHz BS & UE antenna gain 0dB # of UEs in each cell 20 (uniform distribution) # of codes for HS-PDSCH 15 MAX. number of retransmissions 4 HS-SCCH, HS-DPCCH Error free transmission Channel estimation deal Speed of UE 3km/h Roundtrip delay 6 subframe (12ms) Traffic model Open-loop traffic model for HSDPA [8] in which the distribution of packet size is pareto with cut-off α=1.1, k=4.5kbytes and m=2mbytes TABLE MCS LEVELS MCS Modulation Coding rate nfo bit rate level scheme R per code 5 16QAM 3/4 720 kbps 4 16QAM 1/2 480 kbps 3 QPSK 3/4 360 kbps 2 QPSK 1/2 240 kbps 1 QPSK 1/4 120 kbps through the repeated simulations. n Fig. 3, we can find three regions (Region1, Region2, Region3) according to the change of S(,T load ). n Region1, S(,T load ) can have maximum value, 1, since the traffic load is too low. n Region2, the delay criterion is the critical factor when selecting S(,T load ). n Region3, the delay criterion became easy to satisfy, since the delay is larger as traffic is heavier. However, instead of the delay criterion, the fairness criterion is the critical factor when selecting S(,T load ), since the starving or monopolizing problem occurs in the heavy traffic condition. Fig. 4 shows the UE battery power conservation when using the enhanced CQ feedback scheme instead of the periodic CQ feedback scheme. From Fig. 4, we can observe that, with the enhanced CQ feedback scheme, 21% of the battery capacity of UE is conserved compared to the 2ms periodic CQ feedback scheme in view of simulation results, and the amount of UE power conservation decreases as the periodic feedback cycle is longer. n this simulation, we assume that the power consumption of UEs transmitting the CQ value is 1.5 times greater than that of UEs that do not transmit the CQ value.

4 When calculating numerical results, we assume that reading time, packet service time and the priority of UE are fixed and the the distribution of proportional fairness priorities is folded normal distribution. Then, the UE battery power saving when using the enhanced CQ feedback scheme instead of the periodic CQ feedback scheme, P save, can be calculated numerically as follows. P save = MN(,N T S ) t i s P i save(s) (5) where N is the total number of UE in each cell, is the feedback cycle of CQ, T S is packet service time, t i s is service time of UE which has i-th high proportional fairness priority during the feedback cycle of CQ. S(,T load ) is denoted by S and T S is calculated from average packet size and average MCS level found in the simulation. Psave(S) i is the power saving amount due to transmitting fewer CQ feedbacks when UE which has i-th high proportional fairness priority is served. When the scaling factor, S(,T load ) is 1 and the total number of UE is N UE and the periodic feedback cycle is 2ms, Psave(S) i is calculated as follows. Psave(S) i = 1.5 N UE (1.5 (N UE i)+i) (6) 1.5 N UE where 1.5 N UE in equation (6) represents the total power consumption of UEs in the periodic CQ feedback scheme, since the power consumption of UEs transmitting the CQ value is 1.5 times greater than that of UEs that do not transmit the CQ value and every UE transmits the CQ value in the 2ms periodic CQ feedback scheme. However, in the enhanced CQ feedback scheme, if the UE which has i-th high proportional fairness priority is served, the number of UEs transmitting the CQ value is N UE i. Hence, 1.5 (N UE i) in equation (6) represents the power consumption of UEs transmitting the CQ value and i represents the power consumption of UEs that do not transmit the CQ value. Since we assume the fixed packet service time and the priority of UEs, there is a little difference between the simulation results and numerical results as the traffic becomes heavier. V. UPLNK NTERFERENCE ANALYSS The enhanced CQ feedback scheme brings lower uplink interference because the enhanced CQ feedback scheme filters off the redundant CQ feedbacks, which are reported by the UE that is less likely scheduled for transmission due to low proportional fairness priority. For the uplink interference analysis, we assume a system model that is similar to that employed in [5][6]. The system model is summarized in Table. n this system model, UEs in soft handover (SHO) area and them in non-sho area have different CQ feedback cycles. Under the this system model, the E b / o can be calculated as follows. E b E b R SF = o E b (N UE + N (7) 1) Here, R is data transmission rate, SF is spreading factor and N is the average number of CQ feedbacks that the Node 248 Fig. 3. S(,T load ) according to feedback cycle of CQ and traffic load Fig. 4. UE battery saving when using enhanced CQ feedback scheme instead of periodic CQ feedback scheme B receives in one subframe. n the periodic CQ feedback scheme, N is calculated as follows. N = 0.7 N k N. (8) k2 f we assume that the scaling factor is 0.8, the distribution of proportional fairness priorities is uniform and the cycle of periodic CQ feedback is k3, then N of the enhanced CQ feedback scheme is calculated as follows. N = N 0.2. (9) k3 From (7) to (9), we can summarize the E b / o value of uplink DPCCH or HS-DPCCH as shown in Table V. From the Table V, we can see that the enhanced CQ feedback scheme obtains about 2 db gain when the cycle of periodic CQ feedback is 2ms and 1 db gain when the cycle of periodic CQ feedback is 80ms in view of E b / o compared to the periodic CQ feedback scheme.

5 TABLE SYSTEM MODEL FOR UPLNK NTERFERENCE ANALYSS Number of UEs (N UE ) 100 Percentage of UEs in soft handover (SHO) area 30% CQ feedback cycle of UEs in non-sho area (k1) 1(2ms) CQ feedback cycle of UEs in SHO area (k2) 40 (80 ms) HS-DPCCH/UL DPCCH power ratio 0dB Spreading factor of HS-DPCCH/UL DPCCH (SF) 256 Other cell interference Negligible TABLE V EB/O COMPARSON k2 & k3 Eb/o of Eb/o of periodic CQ feedback enhanced CQ feedback 40 (80ms) db db 10 (20ms) db db 1(2ms) db db V. CONCLUSONS n this paper, we proposed the enhanced CQ feedback scheme. n the enhanced CQ feedback scheme, a dynamic threshold is used to filter off the redundant CQ feedbacks, which are reported by the UE that is less likely scheduled for transmission due to low proportional fairness priority. The threshold is calculated by Node B according to the CQ feedback cycle and traffic load. With the enhanced CQ feedback scheme, the battery capacity of UE is conserved and uplink interference is lowered, maintaining the performance of the periodic CQ feedback scheme in view of throughput, delay and fairness. REFERENCES [1] A. Jalali, R. Padovani, R.Pankaj, Data Throughpt of CDMA-HDR a High Efficiency-High Data Rate, Proc. EEE VTC2000-Spring, pp , May [2] 3GPP, Physical layer procedures (FDD), 3GPP TS V [3] 3GPP, Physical channels and mapping of transport channels onto physical channels (FDD), 3GPP TS V [4] 3GPP, Radio Resource Control (RRC): Protocol Specificaition, 3GPP TS V [5] 3GPP, HSDPA Enhancements, 3GPP TS V [6] N. Fukui, Study of Channel Quality Feedback in UMTS HSDPA, PMRC th EEE Proceedings on, vol. 1, pp , Sep [7] R. Jain., The art of computer systems performance analysis, John Wiley and Sons, [8] 3GPP, Physical layer aspects of UTRA High Speed Downlink Packet Access, 3GPP TS V [9] M. Dottling, J. Michel, B. Raaf, Hybrid ARQ and adaptive modulation and coding schemes for high speed downlink packet access, Personal, ndoor and Mobile Radio Communications, 13th EEE nternational Symposium on , vol.3, Sept